OpenAI’s Codex Team Expands Rapidly, Crafting AI That Writes Code With You

OpenAI’s Codex growth has been propelled by two model upgrades released within the last six months. The December 2025 rollout of GPT‑5.2 added a larger context window, allowing the system to retain more project‑level data and reason over it “more effectively than earlier versions,” according to Thibault Sottiaux, head of the Codex group, in an interview with Fast Company. This improvement translated into a noticeable lift in reliability, with the model “working by itself autonomously and reaching really good results,” Sottiaux said. A second boost arrived on February 5, 2026 with GPT‑5.3‑Codex, which Fast Company describes as “substantially improving Codex’s coding chops” and expanding its capacity for multi‑step reasoning, research, and tool use. Developers on X and Reddit quickly reported that the new version could generate usable code on the first try for real‑world projects, a claim echoed by the rapid rise in token consumption—usage measured in tokens has climbed fivefold since the start of the year.

The software‑engineering workflow has shifted from simple autocomplete to what Fast Company calls “agentic engineering.” Codex now operates as an “agentic platform” where multiple autonomous agents can pursue distinct tasks in parallel: one agent hunts for bugs while another drafts documentation, and a third runs integration tests. This orchestration is exposed through the Codex desktop app, launched on February 2, 2026 and marketed as a “command center” for deploying and managing agents. Within weeks, the app surpassed one million downloads, and OpenAI reports that more than half a million users access Codex via the free and “Go” tiers of ChatGPT. The company believes a sizable portion of these users are non‑programmers, since power users tend to gravitate toward higher‑priced plans that offer greater usage limits and faster response times.

Codex’s user base now exceeds one million weekly active developers across all access points—cloud API, desktop app, and command‑line interface—according to OpenAI’s data shared with Fast Company. This tripling of weekly active users mirrors the token surge and reflects broader market adoption of AI‑assisted development. The tool’s ability to converse in plain language, outline project plans, execute code changes, and then explain its reasoning has positioned it as one of the first AI applications delivering measurable business impact, a point Fast Company emphasizes when noting that “generating computer code has emerged as one of the first AI applications making a measurable impact in business.”

OpenAI’s rapid iteration contrasts with competitors such as Anthropic, which introduced Claude Code in the first half of 2025. While both products have moved beyond basic code generation, Fast Company highlights that Codex’s multi‑agent architecture and expanded memory give it a functional edge for large, complex codebases. The company’s internal roadmap, as inferred from Sottiaux’s comments, focuses on further scaling the number of concurrent agents and deepening their ability to integrate external tools—capabilities that could make Codex a de‑facto “assistant engineer” rather than a mere autocomplete service.

The surge in Codex adoption also raises questions about the sustainability of its growth curve. OpenAI’s internal metrics, cited by Fast Company, show that the majority of new users are on lower‑tier subscriptions, suggesting that conversion to higher‑margin plans will be a critical lever for monetization. Moreover, the platform’s reliance on increasingly large models—GPT‑5.2 and GPT‑5.3—implies higher compute costs, a factor that will pressure the company to balance performance gains with operational efficiency. As the AI‑coding market matures, the next competitive frontier will likely be the seamless integration of these agentic workflows into existing development environments, a challenge that OpenAI appears to be tackling head‑on with its Codex desktop “command center.”